Commonly used normalization approaches for quantitative reverse transcription polymerase chain reaction analyses include (a) input nucleic acids standardization (ΔC (q) method), (b) normalizing target gene transcript abundance against a single internal reference gene (ΔΔC (q) method), and (c) geometric averaging of multiple reference gene abundance using the geNorm software. We compared these three approaches to examine expression of a negative muscle growth regulator gene, myostatin-I (mstn-I), in the finfish Lates calcarifer, following 4 weeks of nutritional fasting. Interestingly, these three different approaches led to widely divergent data interpretations. When ΔC (q) and subsequently ΔΔC (q) with α-tub as the reference gene were applied to mstn-I expression data, an ∼threefold upregulation of this gene was observed in fasted compared to fed fish. In contrast, mstn-I appeared unchanged when data was normalized against the geometric average of the two apparently most "stable" reference genes (elongation factor-1 α (ef1-α) and rpl8) selected from a panel comprising seven commonly utilized candidate reference genes (18S, cat-D, ef1-α, rpl8, gapdh, ubq, and α-tub). The geNorm software erroneously suggested that ef1-α, rpl8, and ubq were the most "stable" reference genes, whereas ΔC (q) analysis revealed these genes simply to exhibit similar patterns of regulation in response to fasting. The ΔC (q) approach showed that α-tub was the least variable in its expression level between fasted and fed fish after 4 weeks. The present study also highlights the importance of validating internal references for each time point under investigation when applying ΔΔC (q) and suggests that the more cost-effective ΔC (q) normalization approach, if carefully applied, may in fact produce the most biologically valid results.